Freshwater ecosystems contribute significantly to global methane (CH₄) emissions, yet the role of rivers and streams remains unclear. Previous estimates of global fluvial CH₄ emissions have varied greatly, hampered by the high spatial and temporal variability of CH₄ fluxes in these systems, and by the challenges of measuring both diffusive and ebullitive emissions. Rivers and streams are unique in their ability to produce CH₄ internally while also transporting CH₄ from adjacent wetlands and soils. Therefore, global CH₄ emissions from rivers and streams are likely regulated by multiple environmental factors that operate across land-water boundaries. Understanding these controls is crucial for improving predictions of riverine CH₄ emissions and for a better understanding of how these systems process and transport carbon under the influence of climate change and other anthropogenic factors. This study utilizes a large, newly compiled CH₄ database (GRiMeDB) to model CH₄ concentrations globally using random forest machine learning, aiming to resolve uncertainties in global estimates of riverine CH₄ emissions and identify the primary environmental drivers.

Existing literature highlights the significant contribution of freshwater ecosystems to global CH₄ emissions, with estimates suggesting that they are responsible for nearly half of the total. However, the role of rivers and streams within the freshwater component remains uncertain, with previous global estimates showing wide variations and considerable uncertainty. Studies have shown that fluvial ecosystems are complex systems connecting terrestrial, marine, and atmospheric carbon pools. They can produce CH₄ internally but also receive and emit CH₄ generated externally in adjacent soils and wetlands. This emphasizes the need for research that considers the multitude of environmental factors that influence CH₄ production and transport across land-water boundaries. Previous efforts to quantify global riverine CH₄ emissions have been limited by data scarcity and methodological challenges, leading to large discrepancies between bottom-up inventories and top-down estimates. The variability in CH₄ emissions, spanning several orders of magnitude, also contributes to the uncertainty.

This study leverages the Global River Methane database (GRiMeDB), containing over 24,000 observations of CH₄ concentration and over 8,000 observations of CH₄ fluxes. Spatially explicit global estimates of CH₄ emissions from rivers and streams were developed using random forest machine learning models. These models were trained on the GRiMeDB data to predict CH₄ concentrations globally, accounting for seasonal and spatial variability. The models' performance was evaluated using measures such as R². The model outputs were used to identify the main drivers of CH₄ concentrations and fluxes across various regions. The analysis considered a wide range of environmental variables, including climatic factors (temperature, precipitation), biological factors (primary productivity, soil respiration), edaphic factors (soil organic carbon, peatland cover), physical factors (river slope, elevation, gas-transfer velocity), and anthropogenic factors (population density). Diffusive CH₄ emissions were modeled directly, while ebullitive fluxes, due to data limitations, were estimated using a linear relationship observed between diffusive and ebullitive fluxes in the GRiMeDB data. The uncertainty in the global estimate of CH₄ emissions was quantified using Monte Carlo simulations.

The study found that global CH₄ emissions from rivers and streams account for 27.9 (16.7–39.7) Tg CH₄ per year, a significant contribution to the global CH₄ budget, comparable in magnitude to emissions from other freshwater systems such as lakes and reservoirs. Global patterns of CH₄ concentrations and emissions reveal high fluxes in both tropical and high-latitude regions. In the tropics, high emissions are linked to high rates of terrestrial primary production, soil respiration, precipitation, and wetland connectivity. At high latitudes, they are associated with large soil organic carbon stocks, extensive peatland cover, and shallow groundwater tables. The most influential variables in the models reflect the local landscape, with physical catchment characteristics like river slope, elevation, and gas-transfer velocity negatively influencing CH₄ concentrations, likely due to increased gas exchange between water and atmosphere at higher slopes and velocities. Despite clear global-scale patterns, the study acknowledges substantial unexplained variability at finer scales, highlighting the influence of local controls on CH₄ dynamics not captured by the model's relatively coarse spatial resolution. Contrary to expectations, the temperature dependence of diffusive CH₄ emissions in rivers is low (Eₐ = 0.14 eV), significantly less than in other freshwater ecosystems (Eₐ = 0.96 eV). This is attributed to the open nature of running waters, where external inputs significantly influence CH₄ concentrations. Human population density is also positively correlated with CH₄ concentrations, suggesting a contribution from human activities such as impoundments, agriculture, and wastewater discharge. The tropics account for the largest share of global emissions (37%), followed by Arctic and northern boreal areas (17%). Seasonal patterns show high emissions during the open water season at high latitudes.

The findings challenge the traditional understanding of temperature as a first-order control on aquatic CH₄ production, particularly in river systems. The weak temperature dependence highlights the dominance of lateral CH₄ inputs from surrounding landscapes, indicating that the indirect effects of climate change—through alterations in soil and wetland CH₄ production and hydrological connectivity—are more significant than direct effects on aquatic methanogenesis. This suggests that future models should focus on incorporating landscape and hydrological processes to better predict riverine CH₄ emissions. The significant influence of human activities on CH₄ emissions from rivers and streams underscores the need for mitigation strategies to reduce emissions from these modified ecosystems.

This study provides a robust, spatially explicit estimate of global CH₄ emissions from rivers and streams, highlighting their significant contribution to the global CH₄ budget. The findings emphasize the importance of landscape-scale processes and human activities in regulating CH₄ emissions from these ecosystems. The relatively low temperature sensitivity of riverine CH₄ emissions suggests that future climate change impacts will likely be mediated primarily through indirect effects on surrounding landscapes. Further research should focus on improving the representation of ebullitive CH₄ fluxes and incorporating finer-scale spatial variability in models.

The study acknowledges limitations stemming from the coarser spatial resolution of the model, which does not fully capture finer-scale spatial and temporal variability in CH₄ dynamics, particularly related to groundwater inputs and sediment properties. The estimation of ebullitive CH₄ fluxes relies on a limited dataset, and further measurements are needed to improve accuracy. Furthermore, the analysis excluded data from highly modified systems due to limitations in spatial predictors, potentially underestimating the human contribution to riverine CH₄ emissions in highly altered environments.